<![CDATA[The rapid advancement of Generative Artificial Intelligence (AI) technology poses significant challenges in digital forensics, particularly in distinguishing authentic images from synthetic ones. The limitations of current binary detection methods drive the need for a more comprehensive identification system. This study aims to develop an intelligent web-based system for multiclass AI image detection. The proposed approach utilizes a hybrid model integrating a Convolutional Neural Network (EfficientNet-B0) for local spatial feature extraction and a Vision Transformer (ViT-Base) to capture global visual context. Features from both architectures are combined through a feature fusion mechanism and efficiently classified using a Broad Learning System (BLS) to distinguish six categories: real/AI humans, real/AI animals, and real/AI objects. Evaluation conducted on 900 validation images demonstrates that the hybrid model achieves an overall accuracy of 96.56%, with an average precision, recall, and F1-score of 97%. The system is deployed as an interactive web application utilizing a Streamlit interface and a FastAPI backend. Functional testing proves that the platform can process inferences stably and responsively in real-time. In conclusion, the integration of CNN-ViT and BLS offers superior accuracy alongside optimal computational efficiency for practical digital image authentication.]]>
Copyrights © 2026
